US6983346B2 - Reducing tag-ram accesses and accelerating cache operation during cache miss - Google Patents

Reducing tag-ram accesses and accelerating cache operation during cache miss Download PDF

Info

Publication number
US6983346B2
US6983346B2 US10/435,357 US43535703A US6983346B2 US 6983346 B2 US6983346 B2 US 6983346B2 US 43535703 A US43535703 A US 43535703A US 6983346 B2 US6983346 B2 US 6983346B2
Authority
US
United States
Prior art keywords
address
cache
last
miss
hit
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Expired - Lifetime, expires
Application number
US10/435,357
Other languages
English (en)
Other versions
US20040024967A1 (en
Inventor
Jonathan Y. Zhang
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Texas Instruments Inc
Original Assignee
Texas Instruments Inc
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Texas Instruments Inc filed Critical Texas Instruments Inc
Priority to US10/435,357 priority Critical patent/US6983346B2/en
Assigned to TEXAS INSTRUMENTS INCORPORATED reassignment TEXAS INSTRUMENTS INCORPORATED ASSIGNMENT OF ASSIGNORS INTEREST (SEE DOCUMENT FOR DETAILS). Assignors: ZHANG, JONATHAN Y.
Publication of US20040024967A1 publication Critical patent/US20040024967A1/en
Application granted granted Critical
Publication of US6983346B2 publication Critical patent/US6983346B2/en
Adjusted expiration legal-status Critical
Expired - Lifetime legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F12/00—Accessing, addressing or allocating within memory systems or architectures
    • G06F12/02—Addressing or allocation; Relocation
    • G06F12/08—Addressing or allocation; Relocation in hierarchically structured memory systems, e.g. virtual memory systems
    • G06F12/0802—Addressing of a memory level in which the access to the desired data or data block requires associative addressing means, e.g. caches
    • G06F12/0877—Cache access modes
    • G06F12/0884—Parallel mode, e.g. in parallel with main memory or CPU
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F12/00—Accessing, addressing or allocating within memory systems or architectures
    • G06F12/02—Addressing or allocation; Relocation
    • G06F12/08—Addressing or allocation; Relocation in hierarchically structured memory systems, e.g. virtual memory systems
    • G06F12/0802—Addressing of a memory level in which the access to the desired data or data block requires associative addressing means, e.g. caches
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F12/00—Accessing, addressing or allocating within memory systems or architectures
    • G06F12/02—Addressing or allocation; Relocation
    • G06F12/08—Addressing or allocation; Relocation in hierarchically structured memory systems, e.g. virtual memory systems
    • G06F12/0802—Addressing of a memory level in which the access to the desired data or data block requires associative addressing means, e.g. caches
    • G06F12/0844—Multiple simultaneous or quasi-simultaneous cache accessing
    • G06F12/0855—Overlapped cache accessing, e.g. pipeline
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F12/00—Accessing, addressing or allocating within memory systems or architectures
    • G06F12/02—Addressing or allocation; Relocation
    • G06F12/08—Addressing or allocation; Relocation in hierarchically structured memory systems, e.g. virtual memory systems
    • G06F12/0802—Addressing of a memory level in which the access to the desired data or data block requires associative addressing means, e.g. caches
    • G06F12/0877—Cache access modes
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F12/00—Accessing, addressing or allocating within memory systems or architectures
    • G06F12/02—Addressing or allocation; Relocation
    • G06F12/08—Addressing or allocation; Relocation in hierarchically structured memory systems, e.g. virtual memory systems
    • G06F12/0802—Addressing of a memory level in which the access to the desired data or data block requires associative addressing means, e.g. caches
    • G06F12/0844—Multiple simultaneous or quasi-simultaneous cache accessing
    • G06F12/0855—Overlapped cache accessing, e.g. pipeline
    • G06F12/0859—Overlapped cache accessing, e.g. pipeline with reload from main memory
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F12/00—Accessing, addressing or allocating within memory systems or architectures
    • G06F12/02—Addressing or allocation; Relocation
    • G06F12/08—Addressing or allocation; Relocation in hierarchically structured memory systems, e.g. virtual memory systems
    • G06F12/0802—Addressing of a memory level in which the access to the desired data or data block requires associative addressing means, e.g. caches
    • G06F12/0864—Addressing of a memory level in which the access to the desired data or data block requires associative addressing means, e.g. caches using pseudo-associative means, e.g. set-associative or hashing
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2212/00—Indexing scheme relating to accessing, addressing or allocation within memory systems or architectures
    • G06F2212/10—Providing a specific technical effect
    • G06F2212/1028—Power efficiency
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
    • Y02D10/00—Energy efficient computing, e.g. low power processors, power management or thermal management

Definitions

  • the technical field of this invention is data processor cache.
  • Cache memory is the fundamental block in a hierarchical memory system. Because the entire hierarchy is designated as cache or cache memory, the following terms are adopted here In the text of this application, the lowest level cache designed to handle the most critical data storage is designated as the main cache. Lower levels of cache hierarchy are designated as Level-1 cache and Level-2 cache, higher numbers indicating decreasing order of performance and criticality. Thus, when a processor function requests data, that data is sought from the main cache first. Conventional protocols dictate that the fetch address must be compared to tag addresses stored in a main cache tag-RAM bearing the stored address tags from recent fetch requests. If the fetch address does not match the addresses stored in the tag-RAM then a lower level of cache RAM hierarchy must be queried to obtain the data. These lower level cache RAMs have a similar protocol. No hit for the fetch address results in a corresponding search at yet a lower level of cache memory.
  • the data is typically retrieved from a lower level external memory source and is written into the main cache data-RAM as a fill in anticipation of further need for the data. Future accesses for this same address result in a hit because that the data is available in main cache. If the fetch from cache results in a hit, such data is referred to as resident data already available in the cache.
  • a main cache usually has at least one tag-RAM at least one data-RAM.
  • Stored tag information includes tag valid register bits.
  • the tag-RAM is used to store address tags for the cache lines stored in data-RAM.
  • a cache memory line usually contains more than one entry of data. An entry is a unit of data returned from cache memory to central processing unit (CPU).
  • the tag valid bits indicate whether the tags are valid. If a main cache is an instruction cache, then the CPU can on fetch data from it. If a main cache is a data cache, then the CPU can not only fetch data from it, but also store data to it.
  • This invention can apply equally to instruction cache or data cache, or even lower level caches. An example of main instruction cache is described. Therefore the CPU accesses to the cache consist of only fetches in this example.
  • a fetch to a cache normally requires a tag-RAM. access and a data-RAM access to retrieve the correct data. While a data-RAM access is absolutely necessary to retrieve data from a cache, the tag-RAM access is not entirely unavoidable. Power can be saved if it is possible to avoid access to tag-RAM. The task is to determine how superfluous tag-RAM accesses may be bypassed.
  • miss penalty is defined as the performance impact resulting from a cache miss.
  • miss penalty the performance impact resulting from a cache miss.
  • the latency required to fully stock up a missed cache line ranges from tens to thousands of clock cycles depending on the speed of the lower level of the memory hierarchy and the cache line size. A number of design options have been explored to reduce the waiting period.
  • FIG. 1 illustrates the logical organization of an example prior art two-way cache memory.
  • FIG. 1 illustrates this example the cache having 512 sets identified by reference numbers 100 to 105 . Each set has twoways of data 110 and 111 , tag bits 106 and 107 and valid bits 108 and 109 .
  • Cache line replacement depends upon the least-recently-used (LRU) word 112 , but is not a crucial concern here. What is important in this example is that each cache line has four entries of data in each way. These are entries 120 to 123 in way 110 and entries 130 to 133 in way 111 . The CPU can fetch only one entry each system clock cycle.
  • the cache example of FIG. 1 illustrates the possible advantages that improvements could provide.
  • FIG. 2 illustrates the conventional operation of the example main cache memory of FIG. 1 .
  • data FD 1 corresponding to fetch address FA 1 is initially not resident in the main cache data-RAM 215 .
  • Data FD 2 , FD 3 , FD 4 are also not resident.
  • data FD 5 and FD 6 are resident and available in the main cache data-RAM 215 .
  • a stream of four fetch request addresses 201 including addresses FA 1 , FA 2 , FA 3 , and FA 4 belonging to one cache line, arrive at the input to the main cache memory.
  • Another stream of two fetch request addresses 202 including addresses FA 5 and FA 6 belonging to another cache line, follows.
  • the input fetches will be processed in order of their appearance in the input stream: FA 1 , FA 2 , FA 3 , FA 4 , FA 5 , and FA 6 .
  • brackets [ ] around the addresses are used in FIG. 2 to emphasize the concept.
  • fetch addresses [FA 1 , FA 2 , FA 3 , FA 4 ] of stream 201 belong to one cache line and fetch addresses [FA 5 , FA 6 ] of stream 202 belong to another cache line.
  • a lookup 203 in tag-RAM 210 is conducted. These yield two corresponding tags 204 , 206 and two corresponding valid bits (LV) 205 , 207 for each lookup.
  • Tag compare and validate processing block 208 and cache hit or cache miss query block 213 are performed after each lookup 203 to determine if the wanted data is in the main cache resulting in a cache-hit 216 , or not resulting in a cache-miss 217 .
  • Fetch address FA 1 generates a cache-miss 217 , therefore the CPU fetch operation comes to a halt.
  • the data from the entire cache line including FD 1 , FD 2 , FD 3 and FD 4 has to be transferred from external memory via paths 218 and 226 to cache data-RAM 215 .
  • One of the cache lines in the two-way associative set pointed to by fetch address FA 1 is allocated to host the new cache line and the external memory data is transferred this cache line. The time required to carry out the least-recently-used algorithm is not the main concern of this invention.
  • the tag corresponding to fetch address FA 1 is written to the tag space of the allocated line, and the corresponding line-valid-bit is set, before fetches from the CPU can resume.
  • the fetch data FD 1 can be forwarded directly to the CPU via path 227 .
  • fetches will resume for fetch addresses FA 2 , FA 3 , FA 4 , FA 5 and FA 6 .
  • These five additional fetches are subjected to lookup 203 from the tag-RAM 210 and sequentially generate cache hits 216 .
  • the requested data FD 2 , FD 3 , FD 4 , and FD 5 and FD 6 are read out via signal 216 from data-RAM 215 directly and sequentially via path 228 and supplied to the CPU via path 222 .
  • fetch addresses FA 1 , FA 2 , FA 3 and FD 4 share one tag
  • fetch addresses FA 5 and FA 6 share another tag but the flow causes six tag lookups. Only two tag lookups not six are actually necessary. This difference allows for a first conceptual improvement.
  • This invention is a cache memory employing a tag bypass controller to detect a memory access to the same cache line as a last cache miss address and a last cache hit address. This information is uses for efficient data accesses and forwarding. Registers store the last miss-address and the last hit-address and corresponding valid flags. These hardware features allow reduced tag-RAM accesses and greatly reduce the latency required to fully re-stock a missed cache line.
  • FIG. 1 illustrates the logical organization of an example two-way cache memory (Prior Art);
  • FIG. 2 illustrates the conventional operation of the example two-way cache memory of FIG. 1 (Prior Art);
  • FIG. 3 illustrates a flow diagram of the improved operation of the two-way cache memory of this invention.
  • FIG. 4 illustrates a hardware block diagram of the improved the two-way cache memory of this invention.
  • the first objective of the present invention is to eliminate unnecessary tag-RAM accesses in the main cache. This involves determining under what conditions and in what manner a tag-RAM access may be bypassed.
  • the second objective is to reduce performance impact resulting in a so called miss penalty, which is defined as the performance impact resulted from a cache miss.
  • miss penalty which is defined as the performance impact resulted from a cache miss.
  • the crucial data has to be retrieved from one level lower in a hierarchical memory system. Any reduction of the latency resulting from returning data to the CPU due to the requirement of full restoration of a cache line in the cache (also known as stock-up or restock) provides significant Performance improvement.
  • main cache memory car benefit by two clear improvements that may be identified.
  • a fetch to a main cache normally requires a tag-RAM access and a data-RAM access to retrieve the correct data. While a data-RAM access is absolutely necessary, the tag-RAM access is not entirely unavoidable. It is preferred from power saving point of view not to access tag-RAM if possible. This required determining under what conditions the tag-RAM access may be bypassed.
  • miss penalty is defined as the performance impact resulting from a cache miss.
  • miss penalty the performance impact resulting from a cache miss.
  • the latency required to fully stock up a missed cache line ranges from tens to thousands of clock cycles depending on the speed of the lower level of the memory hierarchy and the cache line size. In order to reduce the waiting period, new design options have been explored.
  • Both the last cache-hit-address and the last cachemissaddress are stored in registers and compared with the current fetch address. If the fetch address belongs to the same cache line as either the cache-hit-address register content or the cache-miss-address register content and the corresponding register contents are valid, then the tag-RAM is bypassed.
  • the cache-hit-address register holds fetch address of the most recent cache-hit.
  • the cache-miss-address register holds the address of the cache-line that is being restocked. This invention does not require that consecutive fetches belong to the same cache line to allow tag-RAM bypass.
  • tag-RAM bypass for this entire fetch stream is still possible.
  • a main cache memory line is in the process of re-stocking, there could be fetches into the main cache that fall into the re-stocking line or a valid main cache line in alternating fashion.
  • the present invention would still enable a tag-RAM bypass even in this case.
  • FIG. 3 illustrates a flow diagram describing the protocol of the improved main cache implementation of this invention.
  • Fetch address inputs 301 or 302 are compared with currently stored cache-miss and cache-hit address registers 316 for match.
  • the address compare is done after truncating a small number of least significant address bits from the cache-miss or cache-hit line addresses and the fetch address.
  • the truncated least significant address bits have a specific value for each separate entry within a cache line.
  • the result of the compare must be validated with the valid flag of a cache-miss address register or a cache-hit address register.
  • a match is declared for miss or hit line address, only if the truncated addresses match with each other and the corresponding valid flag is set.
  • Decision block 307 makes decisions based on the result of comparison of block 306 .
  • a read access 303 is issued to the tag-RAM 310 if a match is detected between the fetch address and either the cache-miss address or the cache-hit address;
  • Tag bypass controller 300 is notified of the result of comparison of block 306 and the decision of decision block 307 .
  • Tag bypass controller 300 monitors the progress of the miss line refill. It is informed which the miss-line entry is received from external memory and which miss-line entry has not been received.
  • Tag bypass controller 300 also remembers the manner in which data may be retrieved from data-RAM for the current cache-hit line and the current cache-miss line.
  • Tag bypass controller 300 takes the following action according to the result of the comparison:
  • tag bypass controller 300 will check whether the fetch data entry has been received.
  • tag bypass controller 300 issues a read 326 to data-RAM to retrieve data.
  • tag bypass controller 300 If no match is found with either cache-hit address or the cache-miss address, then tag bypass controller 300 performs no function for collecting the fetch data. Instead, tag bypass controller 300 waits for the outcome of the tag-RAM access and changes either the cache-miss address or the cache-hit address via path 331 .
  • tag bypass controller 300 will load the fetch address into the cache-hit address register and set a hit valid flag (HVFlag) via path 331 .
  • the data entry will be read out from data-RAM 315 through the normal channels starting from a read request 314 and data output from data-RAM 315 via path 328 .
  • Tag bypass controller 300 will reset HVFlag. Then:
  • the cache stops accepting new fetch commands if a tag compare miss is detected.
  • the CPU pipeline stalls as a result.
  • the cache starts accepting new fetch commands when the data-entry for the missed fetch is received from external memory.
  • the pipeline stalls when address FA 1 is a cache miss. After data FD 1 is supplied to the CPU, the pipeline re-starts. Other entries in the same cache line may still be waiting to be received from external memory when new fetch commands are accepted.
  • Tag bypass controller 300 operates according to case (ii) where a match was found with the cache-hit address but not with the cache-miss address.
  • Tag bypass controller 300 issues a read 326 to data-RAM 315 to retrieve the data.
  • a No response 303 to query 307 causes a normal tag lookup 303 .
  • a Yes response 309 to query 307 causes tag-bypass controller 300 to either initiate tag-bypassed read 326 of data stored in data-RAM 315 or wait for data to be returned from external memory and then forward this data directly to the CPU.
  • FIG. 4 illustrates the additional hardware elements employed by this invention. These are:
  • Cache-hit address register 401 storing the fetch address upon detection of a main cache hit
  • Cache-miss address register 403 storing the fetch address upon detection of a main cache miss
  • Same-line-address detector 406 compares the fetch address with both cache-hit address register 401 and cache-miss address register 403 provided the respective valid bits are set. If either comparison finds the respective compared addresses in the same cache line, a match is declared. If neither comparator finds same-line addresses, then a mismatch is declared. p 4) Address pipeline registers 420 and data-RAM access channel 408 initiate a read from data-RAM 415 when tag-RAM 410 is bypassed.
  • Tag-bypass-controller 400 monitors the return data stream from external memory and makes the correct data forwarding decisions. It also coordinates the access to data-RAM access channel 408 .
  • Cache-hit valid flag 402 signals that the data stored in cache-hit address register 401 is valid. In case of a cache flush or a line replacement, this valid flag will be invalidated.
  • Cache-miss valid flag 404 signals the data stored in cache-miss address register 403 is valid. When the missed line is fully stocked in cache, this valid flag will be invalidated. At this time the valid bit of the fully-stocked cache line is set and subsequent fetches to this main cache line will result in a cache hit.
  • FIG. 4 illustrates further details of hardware operation described later in the text and identifies these seven added hardware elements. It is helpful to refer to these hardware elements in the description of the flow diagram of FIG. 3 applied to a specific example of fetch addresses requested that follows.
  • FIG. 3 places brackets around the addresses included in the same cache line.
  • [FA 1 , FA 2 , FA 3 , FA 4 ] belong to one cache line
  • [FA 5 , FA 6 ] belong to another cache line.
  • a look-up 303 in tag-RAM 310 may be conducted. Each look-up yields two corresponding tags 334 and 336 and two corresponding valid bits (LV) 335 and 337 in each set. The following describes how the two crucial improvements are made to vastly improve upon conventional implementations.
  • fetch address FA 1 When fetch address FA 1 first arrives, same-line detector 306 does not detect a match with either the stored cache-miss address or the stored cache-hit address. Thus query block 307 produces a No result 303 .
  • fetch address FA 1 is stored in cache-miss address register 316 and the miss valid flag is set.
  • Tag look-up 303 is followed by the normal compare and validate 324 and cache hit or miss query 313 .
  • a miss result 313 causes a request for data from external memory 330 to be issued.
  • the cache data line for fetch address FA 1 will be then transferred from external memory to the cache via path 318 and 319 and this data FD 1 corresponding to fetch address FA 1 will be written into data-RAM 315 .
  • Tag-Bypass-Controller 300 is alerted. When the entry FD 2 for FA 2 arrives from external memory via path 318 , then tag-bypass control 300 forwards FD 2 to the CPU directly via control signal 329 .
  • fetch address FA 5 arrives, same line detector 306 compares it with cache-miss address registers 316 and no match is found.
  • the cache-hit address register 316 is not compared because the hit valid flag is not set.
  • Fetch address FA 5 then requires a lookup 303 from tag-RAM 310 and the result is a cache hit.
  • Fetch address FA 5 is stored in cache-hit address register 316 and the hit valid flag is set.
  • the data entry FD 5 already resident in data-RAM 315 is read via read path 326 and sent to the CPU via path 328 .
  • FIG. 4 illustrates the hardware configuration implementing these objectives.
  • Block 417 represents the address request input interface between the CPU and cache, which includes the address bus and fetch requests sent from the CPU.
  • Multiple signal lines 405 represent the fetch address and request signals that are fanned out to blocks 425 , 406 and 420 .
  • Block 410 is the tag-RAM storing tags and valid bits.
  • Block 425 is the logic that merges the address and control signals for reading from and writing to tag-RAM 410 .
  • Signal 432 is the output of tag-RAM 410 , which includes the tags and valid bits for a two-way set that is indexed by tag-RAM read address.
  • Tag compare and validate logic 424 is the hardware that compares the two tags with the fetch address and checks the corresponding valid bits. Since tag-RAM access is triggered by the rising edge of clock, signal 432 becomes available in the trailing clock cycle after fetch address arrives from the CPU. Therefore, address pipeline registers 420 is necessary to register the CPU fetch address to make it available in the same clock cycle when tags and valid bits become available.
  • Tag-RAM write access happens when a cache-miss is detected in tag compare and validate logic 424 .
  • the write access is controlled by tag-RAM write controller 423 .
  • data-RAM access channel 408 is informed via line 409 to read out the corresponding data from data-RAM 415 .
  • the data entry output from 415 is transferred to the data forwarder 407 via path 428 .
  • the data forwarder 407 merges the data-RAM output 428 with data entry from 427 . This data is then sent back to CPU together with a data valid signal via path 442 .
  • Tag-bypass Controller 400 is critical to this invention, because it directs the control signals to carry out the desired flow.
  • Tag-Bypass Controller 400 has the following function:
  • HV Flag Hit-line Valid Flag
  • Miss-line Address Register 403 and Miss-line Valid Flag (MV Flag) 404 are to be updated. If tag compare and validate logic 424 reports a cache-miss and the Miss-line Valid Flag 404 is invalid, the fetch address is stored in the Miss-line Address Register 403 with the MVFlag 404 set to valid. When external data interface 439 reports all entries in the cache line have been transferred to cache, the MVFlag Flag 404 is reset to the invalid state.
  • the Tag Bypass Controller 400 will check whether the required data entry is available in 415 or not. If the data entry is available, the address bus and control signals 421 to data-RAM access channel 408 are activated by tag bypass controller 400 to start a read access. The control signals 419 are sent to data forwarder 407 for proper return of output from data-RAM 415 to CPU. If the data entry is not available in 415 , the CPU fetch will be halted to wait for the data entry to be transferred into cache. Once the required data entry arrives in cache, tag bypass controller 400 will issue a command to data forwarder 407 via 427 to forward the requested data directly to CPU along with the data valid signals. The CPU fetch can be restarted afterwards.
  • the present invention has the capability of reducing tag-RAM lookup frequency by an average factor of four for a cache system with four entries per cache line. As a result, the tag-RAM access power can be reduced by a factor of four. In addition, the present invention improves the latency of the cache. Applications are not limited to the cache configurations illustrated in the examples described. This technique can be generalized for a wide variety of cache configurations.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Memory System Of A Hierarchy Structure (AREA)
US10/435,357 2002-05-10 2003-05-09 Reducing tag-ram accesses and accelerating cache operation during cache miss Expired - Lifetime US6983346B2 (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
US10/435,357 US6983346B2 (en) 2002-05-10 2003-05-09 Reducing tag-ram accesses and accelerating cache operation during cache miss

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
US37950302P 2002-05-10 2002-05-10
US10/435,357 US6983346B2 (en) 2002-05-10 2003-05-09 Reducing tag-ram accesses and accelerating cache operation during cache miss

Publications (2)

Publication Number Publication Date
US20040024967A1 US20040024967A1 (en) 2004-02-05
US6983346B2 true US6983346B2 (en) 2006-01-03

Family

ID=29251251

Family Applications (1)

Application Number Title Priority Date Filing Date
US10/435,357 Expired - Lifetime US6983346B2 (en) 2002-05-10 2003-05-09 Reducing tag-ram accesses and accelerating cache operation during cache miss

Country Status (2)

Country Link
US (1) US6983346B2 (de)
EP (1) EP1361518B1 (de)

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20080040546A1 (en) * 2006-08-14 2008-02-14 Arm Limited. Data processing apparatus and method for performing a cache lookup in an energy efficient manner
US20100223429A1 (en) * 2009-03-02 2010-09-02 International Business Machines Corporation Hybrid Caching Techniques and Garbage Collection Using Hybrid Caching Techniques
US8914580B2 (en) 2010-08-23 2014-12-16 Apple Inc. Reducing cache power consumption for sequential accesses
US20150248354A1 (en) * 2014-03-03 2015-09-03 Kabushiki Kaisha Toshiba Cache memory control circuit and processor

Families Citing this family (11)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US7143243B2 (en) * 2003-05-29 2006-11-28 Via-Cyrix, Inc. Tag array access reduction in a cache memory
US20070005912A1 (en) * 2005-07-01 2007-01-04 Ming-Chung Kao Apparatus and method for reducing tag ram access
US20080010087A1 (en) * 2006-01-30 2008-01-10 Daniel Ronnie C Referral coordination systems and methods
US7523267B2 (en) * 2006-09-15 2009-04-21 International Business Machines Corporation Method for ensuring fairness among requests within a multi-node computer system
JP2008107983A (ja) * 2006-10-24 2008-05-08 Nec Electronics Corp キャッシュメモリ
US8145874B2 (en) * 2008-02-26 2012-03-27 Qualcomm Incorporated System and method of data forwarding within an execution unit
US10402324B2 (en) * 2013-10-31 2019-09-03 Hewlett Packard Enterprise Development Lp Memory access for busy memory by receiving data from cache during said busy period and verifying said data utilizing cache hit bit or cache miss bit
CN103810122B (zh) * 2013-12-24 2017-03-22 苏州国芯科技有限公司 用于嵌入式系统的高速缓存器
CN103810116B (zh) * 2013-12-24 2017-06-20 苏州国芯科技有限公司 用于嵌入式系统的指令存储装置
CN103810117B (zh) * 2013-12-24 2017-03-22 苏州国芯科技有限公司 处理器用指令存储系统
CN118568012B (zh) * 2024-07-29 2024-10-15 北京开源芯片研究院 一种访存请求的处理方法、装置、电子设备及存储介质

Citations (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US4370710A (en) 1980-08-26 1983-01-25 Control Data Corporation Cache memory organization utilizing miss information holding registers to prevent lockup from cache misses
US4905141A (en) * 1988-10-25 1990-02-27 International Business Machines Corporation Partitioned cache memory with partition look-aside table (PLAT) for early partition assignment identification
US5860104A (en) * 1995-08-31 1999-01-12 Advanced Micro Devices, Inc. Data cache which speculatively updates a predicted data cache storage location with store data and subsequently corrects mispredicted updates
US5987561A (en) * 1995-08-31 1999-11-16 Advanced Micro Devices, Inc. Superscalar microprocessor employing a data cache capable of performing store accesses in a single clock cycle
US20010049772A1 (en) 2000-06-05 2001-12-06 Samsung Electronics Co., Ltd. Method and apparatus for reducing power in cache memories and a data processing system having cache memories
US6584549B2 (en) * 2000-12-29 2003-06-24 Intel Corporation System and method for prefetching data into a cache based on miss distance
US20040123034A1 (en) * 2002-12-23 2004-06-24 Rogers Paul L. Multiple cache coherency
US20040153607A1 (en) * 2000-08-07 2004-08-05 Rowlands Joseph B. Direct access mode for a cache
US6775741B2 (en) * 2000-08-21 2004-08-10 Fujitsu Limited Cache system with limited number of tag memory accesses

Patent Citations (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US4370710A (en) 1980-08-26 1983-01-25 Control Data Corporation Cache memory organization utilizing miss information holding registers to prevent lockup from cache misses
US4905141A (en) * 1988-10-25 1990-02-27 International Business Machines Corporation Partitioned cache memory with partition look-aside table (PLAT) for early partition assignment identification
US5860104A (en) * 1995-08-31 1999-01-12 Advanced Micro Devices, Inc. Data cache which speculatively updates a predicted data cache storage location with store data and subsequently corrects mispredicted updates
US5987561A (en) * 1995-08-31 1999-11-16 Advanced Micro Devices, Inc. Superscalar microprocessor employing a data cache capable of performing store accesses in a single clock cycle
US20010049772A1 (en) 2000-06-05 2001-12-06 Samsung Electronics Co., Ltd. Method and apparatus for reducing power in cache memories and a data processing system having cache memories
US6560679B2 (en) * 2000-06-05 2003-05-06 Samsung Electronics Co., Ltd. Method and apparatus for reducing power consumption by skipping second accesses to previously accessed cache lines
US20040153607A1 (en) * 2000-08-07 2004-08-05 Rowlands Joseph B. Direct access mode for a cache
US6775741B2 (en) * 2000-08-21 2004-08-10 Fujitsu Limited Cache system with limited number of tag memory accesses
US6584549B2 (en) * 2000-12-29 2003-06-24 Intel Corporation System and method for prefetching data into a cache based on miss distance
US20040123034A1 (en) * 2002-12-23 2004-06-24 Rogers Paul L. Multiple cache coherency

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
Wilson, K.,M., et al.; Increasing Cache Port Efficiency for Dynamic Superscalar Microprocessors, Proc. of the 23<SUP>rd </SUP>Annual Symp. On Computer Architecture. Philadelphis, PA, May 22-24, 1996, Proc. of the Annual Symp. On Computer Architecture, NY, ACM/IEEE, US, vol. SYMP. 23, May 22, 1996, pp. 147-157.

Cited By (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20080040546A1 (en) * 2006-08-14 2008-02-14 Arm Limited. Data processing apparatus and method for performing a cache lookup in an energy efficient manner
US7529889B2 (en) * 2006-08-14 2009-05-05 Arm Limited Data processing apparatus and method for performing a cache lookup in an energy efficient manner
US20100223429A1 (en) * 2009-03-02 2010-09-02 International Business Machines Corporation Hybrid Caching Techniques and Garbage Collection Using Hybrid Caching Techniques
US8312219B2 (en) 2009-03-02 2012-11-13 International Business Machines Corporation Hybrid caching techniques and garbage collection using hybrid caching techniques
US8738859B2 (en) 2009-03-02 2014-05-27 International Business Machines Corporation Hybrid caching techniques and garbage collection using hybrid caching techniques
US8914580B2 (en) 2010-08-23 2014-12-16 Apple Inc. Reducing cache power consumption for sequential accesses
US20150248354A1 (en) * 2014-03-03 2015-09-03 Kabushiki Kaisha Toshiba Cache memory control circuit and processor
US9715461B2 (en) * 2014-03-03 2017-07-25 Kabushiki Kaisha Toshiba Cache memory control circuit and processor

Also Published As

Publication number Publication date
EP1361518B1 (de) 2013-08-07
EP1361518A1 (de) 2003-11-12
US20040024967A1 (en) 2004-02-05

Similar Documents

Publication Publication Date Title
JP2554449B2 (ja) キャッシュ・メモリを有するデータ処理システム
US7415575B1 (en) Shared cache with client-specific replacement policy
US6983346B2 (en) Reducing tag-ram accesses and accelerating cache operation during cache miss
US6499085B2 (en) Method and system for servicing cache line in response to partial cache line request
US5353426A (en) Cache miss buffer adapted to satisfy read requests to portions of a cache fill in progress without waiting for the cache fill to complete
US7383394B2 (en) Microprocessor, apparatus and method for selective prefetch retire
US6212603B1 (en) Processor with apparatus for tracking prefetch and demand fetch instructions serviced by cache memory
US6678792B2 (en) Fast and accurate cache way selection
US20020010839A1 (en) Multiple variable cache replacement policy
US20080189487A1 (en) Control of cache transactions
JPH06243039A (ja) キャッシュメモリシステムおよびマイクロプロセッサ内の命令を操作するための方法
US6718454B1 (en) Systems and methods for prefetch operations to reduce latency associated with memory access
US6976126B2 (en) Accessing data values in a cache
KR20150016278A (ko) 캐시 및 변환 색인 버퍼를 갖는 데이터 처리장치
WO2002093385A2 (en) Method and system for speculatively invalidating lines in a cache
JPH08272682A (ja) ロード・バッファを備えたロード／ストア・ユニット内にて切り離されたタグ及びデータ・アレイへのアクセス方法及びそのアレイを有する装置
US20040215882A1 (en) Cache predictor for simultaneous multi-threaded processor system supporting multiple translations
US7370152B2 (en) Memory controller with prefetching capability
US9280476B2 (en) Hardware stream prefetcher with dynamically adjustable stride
US5897651A (en) Information handling system including a direct access set associative cache and method for accessing same
JPH10214226A (ja) 第２レベルキャッシュの古いラインの除去によるプロセッサのメモリ性能の強化方法及びシステム
US20110022802A1 (en) Controlling data accesses to hierarchical data stores to retain access order
EP0543487B1 (de) Verfahren und Cache-Speichersteuereinheit zum Holen von Daten für eine Zentraleinheit mit verringerter Wartezeit
US20100146212A1 (en) Accessing a cache memory with reduced power consumption
JP3431878B2 (ja) マルチスレッド・プロセッサの命令キャッシュ

Legal Events

Date Code Title Description
AS Assignment

Owner name: TEXAS INSTRUMENTS INCORPORATED, TEXAS

Free format text: ASSIGNMENT OF ASSIGNORS INTEREST;ASSIGNOR:ZHANG, JONATHAN Y.;REEL/FRAME:014065/0152

Effective date: 20030506

STCF Information on status: patent grant

Free format text: PATENTED CASE

FPAY Fee payment

Year of fee payment: 4

FPAY Fee payment

Year of fee payment: 8

FPAY Fee payment

Year of fee payment: 12